Hurricane Andrews Landfall in South Florida. Part I

Home , Hurricane Alicia, Hurricane Andrew, Hurricane Bob (1979), Hurricane Donna, Hurricane Emily (1987), Hurricane Emily (1993)

304 WEATHER AND FORECASTING VOLUME 11

Hurricane Andrew's Landfall in South Florida. Part I: Standardizing Measurements for Documentation of Surface Wind Fields

MARK D. POWELL AND SAMUEL H. HOUSTON Hurricane Research Division, NOAA/AOML, Miami, Florida

TIMOTHY A. REINHOLD Department of Civil Engineering, Clemson University, Clemson, South Carolina (Manuscript received 10 April 1995, in ®nal form 8 February 1996)

ABSTRACT Hurricane Andrew's landfall in south Florida left a swath of destruction, including many failed anemometer recording systems. Extreme destruction led to exaggerated claims of the range of wind speeds that caused such damage. The authors accumulated all available data from surface platforms at heights ranging from 2 to 60 m and reconnaissance aircraft at altitudes near 3 km. Several procedures were used to represent the various types of wind measurements in a common framework for exposure, measurement height, and averaging period. This set of procedures allowed documentation of Andrew's winds in a manner understandable to both meteorologists and wind engineers. The procedures are accurate to {10% for marine and land observing platforms, and boundary layer model adjustments of ¯ight-level winds to the surface compare to within 20% of the nearest surface measurements. Failure to implement the adjustment procedures may lead to errors of 15%±40%. Quality control of the data is discussed, including treatment of peak wind observations and determination of the radius of maximum winds at the surface.

1. Introduction a life of their own and still persist with the public; many Hurricane Andrew's landfall in south Florida on 24 residents feel that they have survived a storm with August 1992 (May®eld et al. 1994) ranks as the worst winds much worse than actually occurred. natural disaster in U.S. history in terms of destruction A controversy developed after a preliminary report and property loss. Andrew was the ®rst Saf®r±Simpson from the Wind Engineering Research Council (WERC (Simpson and Riehl 1980) category 4 hurricane to 1992) suggested weaker winds than those issued in ad- strike a major urban area in Florida since Donna in visories and a preliminary report by the National Hur- 1960. Hurricanes of similar intensity to Andrew may ricane Center (NHC 1992). Additional reports based become more frequent in the Atlantic basin if climatic on damage investigations and model simulations fueled in¯uences on circulation patterns lead to a return to a the controversy (Wallace 1992). The implications of period of active hurricane formation and landfall these and other wind studies would in¯uence the out- (Landsea 1993). come of millions of dollars worth of litigation related Immediately following Andrew, great interest arose to storm damage. Two sides of the wind speed issue regarding the magnitude and extent of the winds re- emerged: 1) Andrew was an act of God and it is im- sponsible for both the widespread damage and the lo- possible to economically design and build a structure calized swaths of destruction. Unsubstantiated rumors to withstand such winds, and 2) Andrew's winds were circulated in the media of extreme sustained winds no stronger than other recent hurricanes such as Hugo ``clocked'' anywhere from 150 to 214 mph at several in 1989; with good workmanship and proper safety fac- locations (the Pioneer Museum, Turkey Point nuclear tors taken into account, structures should have been power plant, Homestead Air Force Base, and a U.S. able to hold up to the wind loading. The merits of these Army communications facility) where wind records viewpoints are still being argued in the courts. Com- were nonexistent or incomplete. These rumors took on pounding the controversy was the tendency for meteorologists and wind engineers to use different terminology and procedures for dealing with surface wind measurements and analyses. Corresponding author address: Dr. Mark D. Powell, Hurricane The purpose of this paper is to outline a series of Research Division, NOAA/AOML, 4301 Rickenbacker Causeway, Miami, FL 33149. analysis methods for hurricane wind observations that E-mail: [email protected] produce wind ®elds in a common framework for both

᭧ 1996 American Meteorological Society

/3q05 0218 Mp 304 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 305

FIG. 1. Track of Hurricane Andrew's wind center, with locations of surface observations and their survivability. Numbers refer to sources listed in Table 1. Site 26 (not shown) is about 12 km north of site 25.

meteorologists and wind engineers. These methods yses of Andrew's strongest surface winds, wind ®eld conform to accepted practice in the ®elds of micro- decay over south Florida, and potential applications meteorology and wind engineering and design and are for real-time analysis and preliminary damage as- being employed in a surface wind analysis system un- sessment. der development for eventual transfer to NHC. This paper will ®rst discuss failures of anemometer 2. Data availability and anemometer system systems relative to Andrew's track and strategies for problems improving their performance. The turbulent and convective characteristics of hurricane winds will then be The storm track (Fig. 1) represents the track of the considered as suggested by spectral density calcula- circulation center, based on wind measurements from tions from several storms. The importance of a com- U.S. Air Force Reserves reconnaissance aircraft at 3- mon observing framework will be highlighted, stress- km altitude and public reports of calm observed at the ing height adjustment to the reference level of 10 m surface; positions of wind measurement sites are also and its dependence on roughness and stability over land plotted. Wind measurement systems failed for a va- and offshore. Adjustment to standard terrain (open ex- riety of reasons (Table 1) including lack of recording posure) over land follows, together with the determi- capability and mast, crossarm, and guy wire failures nation of the maximum 1-min sustained wind speed caused by a combination of design faults, high winds, based on gust factor data and averaging period. Finally, and debris. Most sites were installed for a speci®c pur- quality control of the Andrew dataset is described, in- pose: aviation, air quality, agriculture, general inter- cluding consideration of the placement of adjusted re- est. In our opinion, very few sites would have failed connaissance wind observations. The companion paper if they were installed to survive hurricane episodes. (Powell and Houston 1996) describes objective anal- The fact that several instruments survived that lacked

/3q05 0218 Mp 305 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 306 WEATHER AND FORECASTING VOLUME 11

TABLE 1. Anemometer survival in Hurricane Andrew: complete, partial, and failed sources.

Size Comments

Station failed or dismantled, record unavailable or unusable (14) 1. NOAA/NWS/MIA Electrical problem station out before storm 2. FAA/LLWAS One station performed erratically 3. NOAA/AOML Dismantled to protect atrium windows 4. UM/RSMAS Dismantled to protect sensors used for climate program 5. UM/engineering Inadequate information on sampling 6. Metro Dade/emergency operations center No recording capability, log unavailable 7. Metro ®re/rescue No recording capability, log unavailable 8. Army communication site No recording capability, no log 9±11. Agricultural extension service Three sites had central datalogger waterlogged and shorted 12. Metro Water and Sewer pump 5 Mast failed, chart record timing error, calibration error 13. Homestead AFB Power turned off to wind recorder before storm 14. Ocean Reef emergency operations center Station dismantled for maintenance

Station failed with partial record (10) 15. NOAA/NHC (NHC) Crossarm rotated, dropped sensor to roof 16. Seaward explorer No recording capability, manual obs 17. FAA/Tamiami Airport No recording, manual obs, crossarm rotated, mast failed 18. Perrine homeowner (PER) No recording capability, manual obs, mast failed 19. Fowey Rocks C-MAN Mast failed at base attachment after 0800 UTC 20. FPL 1 Turkey Point Mast failed, cups blown out of frames 21, 22. FPL 2 60-m mast (two sensors) Cups blew out of frames, no backup power 23. Agricultural Extension Service No obs transmitted after 0730 UTC (one ob only at Glider Port) 24. Metro Water and Sewer at Virginia Key 1

Station survived with complete record (10) 25. NOAA/Miami Beach DARDC 26. NOAA/NOS Next-Generation Water Level site at Haulover Pier 27. Metro Water and Sewer at Virginia Key 2 28±31. FAA/LLWAS (four sites) at Miami International Airport 32. Sailing yacht Mara Cu, North Key Largo 33. National Park Service station at Manatee Bay 34. National Park Service station at Joe Bay

recording capability is disturbing. ``Eyewitness'' ob- After a recommendation by the Interdepartmental servations from the digital or analog display of sites Hurricane Conference (OFCM 1993a), a retro®tting without recorders were extremely dif®cult to docu- kit was developed to lock the crossarm on similar sys- ment and con®rm. In particular, determining the time tems. In addition, the Automated Surface Observing and duration of extreme winds observed in this man- System (ASOS) anemometer design is also being ner was especially dif®cult. studied for similar failure potential. An example of one type of failure suffered by an- The highest surface winds measured in the storm emometers installed by the Federal Aviation Admin- came from anemometers (Perrine and Fowey Rocks C- istration (FAA) and by the National Weather Service MAN) that experienced mast failures shortly after- (NWS) is shown in Fig. 2. This type of failure oc- ward. As discussed below, both measurements occurred at Tamiami Airport and at NHC. Conduit pipe curred in the northern eyewall, in close proximity to used for the crossarm unscrewed itself when subjected the maximum 10-s ¯ight-level winds measured by the to torque produced by the drag force on the anemom- air force reconnaissance aircraft. Although it is possible eter. Once rotated past 90 degrees, the relatively heavy that slightly higher winds occurred after the failures, weight of the instrument contributed to pull the in- we believe that these measurements adequately sam- strument out of its mounting socket and onto the pled Andrew's maximum sustained winds and that sub- ground. Afterward, the mast was hit by debris visible sequent winds were not signi®cantly higher. in Fig. 2. The added drag on the mast pulled one of The problems with wind data availability summa- the guy wires out of the ground, causing the mast to rized in Table 1 are not unique to Andrew; these dif- topple. Unfortunately, this site had no recording ca- ®culties plague all major tropical cyclone landfalls pability. The mast at NHC did not topple, but the an- worldwide. The most promising strategies for improv- emometer fell off the crossarm in an identical manner. ing the availability of hurricane wind measurements

/3q05 0218 Mp 306 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 307

FIG. 2. F-420 anemometer failure at Tamiami Airport. Note rotation of crossarm (shown by arrow), which allowed anemometer to be twisted off its mounting. include the following: 1) Incorporate recording capa- 3. Hurricane wind characteristics bility on all current anemometer systems and inspect a. Spectral density instrument masts for survivability. 2) Adopt a surface wind measurement standard containing guidelines for If meteorologists and wind engineers hope to ade- survivable anemometer performance characteristics, quately measure the wind ®eld of landfalling tropical siting, emergency power, and archival recording that is cyclones, it is imperative that we learn as much as pos- consistent with recommendations of the World Mete- sible about processes that in¯uence the ¯ow ®eld and orological Organization (WMO). A draft standard has its temporal and spatial scales. Typically, the only in- been submitted to the American Society for Testing of formation available to study such processes has been Materials for consideration as a standard (T. Lockhart, strip chart records from landfalling hurricanes and ty- Meteorological Standards Institute, 1994, personal phoons, and these records suffer from poorly docu- communication), based in part on reaction to limita- mented response characteristics of the chart±sensor tions of the ASOS system (Powell 1993). 3) Support system. A few digital time series were collected by of a university/federal agency/private industry net- Brookhaven National Laboratory (BNL) on Long Is- work of preidenti®ed anemometer sites that can either land, New York, for several landfalling hurricanes in be instrumented each hurricane season or rapidly de- the 1950s and 1960s, by the Ocean Current Measure- ployed locally when a hurricane is approaching. This ment Program for two hurricanes in the Gulf of Mex- deployment approach has the advantage of local logis- ico, and by the Field Research Facility of the U.S. tics to allow enough time for proper installation of a Army Corps of Engineers (USACE) Coastal Engi- measurement site. neering Research Center at Duck, North Carolina, for

/3q05 0218 Mp 307 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 308 WEATHER AND FORECASTING VOLUME 11

Hurricane Bob in 1991. Unfortunately, no high-resolution time series data of this type were recorded in Hurricane Andrew, but the spectral density characteristics from these other datasets may shed some light on the scales of motion that may have been present. BNL tower observations were collected in Hurri- canes Carol (1954), Edna (1954), Connie (1955), and Donna (1960) at Ç100 m as described in van der Hoven (1957), Davenport (1961), and Singer et al. (1961). Mean wind observations in these storms ranged from 20 to 29 m s 01 . Van der Hoven (1957) used the Connie observations together with longer, nonhurricane time series records to establish the exis- tence of a mesoscale spectral gap at periods corresponding to 0.1±5.0 h. Davenport (1961) examined several high-wind datasets, including Hurricanes Carol and Edna. His spectral density plots were normalized FIG. 3. Spectral density plot of detrended ¯uctuations of the by the estimated surface stress. All the BNL spectra streamwise component of the wind at 20 m measured at the USACE Duck, North Carolina, pier for alongshore ¯ow in 23 m s01 mean show a hump shape beginning with a magnitude of 0.5 winds during Hurricane Bob on 19 August 1991. Spectral estimates m2 s 02 at a frequency of 2 h 01 , reaching a peaks of 3± have been smoothed with a 20-point Hanning ®lter. Vertical axis is 20 m2 s 02 at 50 h 01 (1.2 min) and then trailing off at the product of frequency and spectral density, which has units of higher frequencies of 10 3 h 01 . BNL is situated in an variance; horizontal axis is frequency expressed in cycles per hour on a logarithmic scale. The vertical line with arrows refers to the 95% open ®eld amid a scrub pine forest, resulting in a rough- con®dence interval applied to an estimate of 0.1 m2 s02. The area ness length of 1 m. Forristall (1988) analyzed wind beneath the curve is proportional to the contribution of a given fre- spectra collected from offshore oil rig platforms in the quency band to the total variance. Gulf of Mexico during Hurricanes Eloise (1975) and Frederic (1979) in mean winds of 29 m s 01 at the 40- m (Eloise) and 70-m (Frederic) levels. His spectral b. Applicability of similarity theory density plots were normalized by the wind variance and A method is required to adjust winds to a common showed a similar shape to those from BNL over a fre- height, exposure, and averaging time. Height and ex- quency band from 0.3 to 3600 h 01 with a peak of 20%± posure adjustment procedures depend on relationships 30% of the variance (or Ç3m2s02)at30±300h01 developed from Monin±Obukov similarity theory. Ac- (0.2±2.0 min). cording to Panofsky and Dutton (1984) and Arya Spectral density analysis of Hurricane Bob (Fig. 3) 01 (1988), similarity requires a horizontally homoge- in a mean alongshore ¯ow of Ç23 m s (at 20-m neous surface layer where the mean ¯ow and turbulent level) show agreement with previous studies in the characteristics depend only on height, surface stress, shape of the spectrum but display much lower peak 2 02 01 surface heat ¯ux, and buoyancy. The ¯ow is assumed energy levels (0.2±0.3 m s ) at 20±300 h (0.2± to be quasi-stationary, turbulent ¯uxes of momentum 3.0 min) in accordance with lower wind speeds and and heat are assumed independent of height in the sur- smoother terrain. All the spectra are dominated by tur- face layer, and molecular and rotational effects are neg- bulent (100±2000 h 01 , 2±30 s) and convective me- 01 ligible. Hence, the wind direction is constant within the sogamma (10±100 h , 0.5±6.0 min) scales, with no surface layer, and the variation of wind speed with evidence of larger mesoscale phenomena. In lower height is controlled by the surface stress, terrain rough- speeds, however, the Hurricane Bob data (not shown) ness, and surface heat ¯ux. To the extent that turbulent show evidence of considerable energy in mesogamma and smaller convective scales are important, the char- 01 and mesobeta scales (0.5±10 h , 6±30 min) that may acteristics of the ¯ow should be well represented by be associated with rainbands or boundary layer roll vor- surface-layer similarity theory. If the ¯uctuations of the tices (Tuttle and Gall 1995). The effects of rainbands ¯ow are primarily affected by cloud- and rainband- and individual convective elements on spectral shape scale phenomena, surface-layer similarity may fail to would probably also be dependent on the track of the consider important scaling parameters. In a hurricane, storm relative to the measurement site. In Andrew, with rotational effects associated with curved ¯ow may also surface winds of 40±60 m s 01 , it is likely that the peak be important. The motion of a storm makes it unlikely in the spectrum would be found at the turbulent and that quasi-stationarity can be satis®ed except for short convective-mesogamma scales. Applying Davenport's time periods in slow-moving storms. Even though the (1961) normalized spectrum shape to Andrew would mean ¯ow may be changing with time, the distribution yield peak energy for open terrain on the order of 10± of short-period means about longer-period means may 20 m2 s 2 . not change greatly from one time period to the next.

/3q05 0218 Mp 308 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 309

For example, histograms of deviations of the 1-min corded a dense collection of wind measurements from mean wind about the 10-min (Fig. 4a) and 1-h means which spatial variability of the wind ®eld could be stud- (Fig. 4b) from data collected in Hurricane Bob suggest ied. One-minute averages computed from these data for a Gaussian distribution. Closer to the eyewall, or 0935 UTC 24 August are displayed in Fig. 6a, along for a faster-moving storm like Andrew, considerable with anemometer heights ZAN , estimated roughness changes can occur over 1 h, but periods of order 10 lengths ZO , and zero-plane displacement heights ZD min would be relatively stable. (described below) based on site visits and photo- Despite these inadequacies, similarity-based proce- graphic descriptions of the site exposures. The variance dures and boundary layer adjustment models appear to of the wind observations displayed in Fig. 6a is con- work reasonably well and have been used to adjust siderable (19.6 m2 s 02 ) with the wind speeds varying ¯ight-level reconnaissance measurements to the sur- by more than a factor of 2 between the lowest and high- face (Powell 1980) and also to reconstruct hurricane est observations. These differences are due to many wind ®elds at landfall (Powell 1982, 1987; Powell et factors, including anemometer height, exposure differ- al. 1991). In this paper, subject to the limitations dis- ences, and the natural variability of the wind associated cussed above, we will assume that 1) similarity theory with turbulent eddies and convective cells over a 50 relationships are applicable for height adjustment of km2 area. After averaging the data for a longer period measured winds and 2) gust factor relationships de- (10 min), adjusting the observations to a common veloped under the Gaussian assumption are applicable height of 10 m, and further adjusting the observations for estimating short period wind extremes from longer to a common exposure typical of an airport runway period averages, or vice versa. (Fig. 6b), the variance is reduced to 3.6 m2 s 02 , a reduction of 79%. 4. Standardization of Hurricane Andrew wind A closer look at the exposures of these sites helps to measurements account for the differences; site FA1 (Fig. 7a) was the open exposure site in the middle of the runway grid, To analyze the surface wind ®eld accurately, it is sites FA2 and FA4 were in suburban terrain north and critical that input data conform to a common height, south of the airport, and site FA3 (Fig. 7b) was exposed averaging period, and exposure. Wind observations un- very poorly to the east of the airport. Without infor- corrected for height, fetch, and time period may show mation on the instrument exposure and height, one large differences over short distances (caused by ter- might expect the sites to be comparable and conclude rain roughness and associated turbulence) that mask that the observed differences in wind speed were larger-scale wind features associated with the storm. caused by some small-scale feature associated with the Objective analysis ®lters can smooth out these differ- storm. As weather services are modernized, it is hoped ences, but unrepresentative information will then be- that all wind sites will eventually be catalogued for come part of the analyzed ®eld. Our approach to the exposure and sensor height, allowing standardization reconstruction of Andrew's wind ®eld is to standardize of measurements to an exposure representative of a de- input observations to a common reference framework sired analysis product or consistent with numerical before analysis. The standardization generally requires weather prediction model land use and roughness spec- a ®ve-step process (Fig. 5): 1) compute the mean (10 i®cations. min) surface wind from continuous short-period means (if available) or extreme short-period means (by use a. Determination of exposure of gust factor relationships), 2) determine the observation exposure type, 3) adjust to the reference height In most cases the upwind fetch of a wind observation (10 m), 4) adjust to the reference exposure, and 5) can be roughly categorized as either ``land'' or ``ma- adjust to the maximum sustained (1-min mean) wind rine.'' For coastal stations, stations just inland or just over the averaging period. Step 1 is required because offshore, the fetch depends upon whether the wind di- the surface-layer similarity relationships used in steps rection is on-, off-, or alongshore. Winds proceeding 2±4 require a mean wind and the 10-min mean is a from sea to land decelerate and form an internal bound- more stable measurement of the mean wind than a 1- ary layer (IBL, Fig. 8); winds from land to sea accel- min mean. These steps also apply to determination of erate. For neutral and for unstable conditions, the peak winds for engineering applications, except that height of the IBL may be approximated (Peterson step 5 refers to a peak 3-s wind speed (ASCE 1994). 1969; Wood 1982; Arya 1988) by Each step is documented below along with estimates of the uncertainty and consequence error. X 0.8 hIORÅ cZ , (1) As an example, consider the FAA Low Level Wind- ͩͪZOR shear Alert System (LLWAS) measurements collected in the vicinity of Miami International Airport, about 36 where c is a constant dependent on stability with values km north of Andrew's wind center. This location was between 0.28 and 0.75, ZOR is the larger of the two not affected by Andrew's eyewall, but LLWAS re- roughness lengths, and X is the distance downwind of

/3q05 0218 Mp 309 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 310 WEATHER AND FORECASTING VOLUME 11

FIG. 4. Histogram of the deviations of 1-min means about (a) 10-min means and (b) 1-h means observed during Hurricane Bob at the USACE Duck, North Carolina, pier on 19 August 1991. Curves represent ®ts of a Gaussian distribution to the data. The vertical axis refers to the number of observations within a 0.42 m s01 (left) or 0.48 m s01 (right) bin.

the roughness change. Winds above hI are the same as The failure of the ¯ow to reach equilibrium was read- would be measured over the upwind roughness. Winds ily apparent when comparing observations from the below hE Å 0.1hI are in equilibrium with the new LLWAS system. The extreme north and south stations, roughness, and between hE and hI , the wind is in the FA2 and FA4, were both in suburban terrain with process of adjusting to the new underlying roughness. roughness lengths estimated at 0.5±1.0 m based on the By estimating ZOR as described below, measuring X guidelines of Wieringa (1992). These estimates per- from a map or navigation chart, and comparing hE to formed well in adjusting anemometer heights to 10 m the ZAN of the station of interest, one can estimate the and open-terrain exposure (as de®ned by comparison exposure category according to whether ZAN is closer to FA1 observations) when wind directions were from to hI (same category as the upwind roughness) or hE the north (overland fetch). However, when the wind (same category as the downwind roughness). veered to the northeast, open-terrain-adjusted winds at Onshore air¯ow can experience several additional FA2 and FA4 (using the assigned roughness lengths) IBL formations depending on land use and topography. failed to compare with FA1. Apparently, the distant In such conditions, a wind pro®le with suf®ciently long marine fetch (even at 20±30 km upwind) became im- averaging times might show kinks associated with sep- portant, and roughness estimates had to be reduced arate hI values or a new transitional boundary layer may (Fig. 6) to produce good comparisons to FA1. This be evident that has not reached equilibrium with the behavior of higher-estimated versus lower-measured new underlying surface. Furthermore, in the early hours (from wind pro®le) roughness was observed for ¯ow of an approaching hurricane or the late hours following from sea to land for fetches up to 45 km inland in Den- landfall, the wind direction may parallel the coastline. mark by Sempreviva et al. (1990). Models attempting In this situation it is dif®cult to characterize the expo- to deal with multiple IBL formation along a downwind sure because transverse eddies in the upstream ¯ow trajectory have been developed by Larsen et al. (1982). ®eld will cause the wind to have characteristics of ei- Conceivably, such models could predict downwind ther land or marine or a combined exposure. ¯ow based on a knowledge of roughness lengths along

/3q05 0218 Mp 310 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 311

cussed below; wind directions are assumed to be independent of height in the surface layer.

1) LAND EXPOSURE

Mean surface wind measurements Uz at a height Z ú ZD m above the surface are adjusted to the 10-m- level wind speed U10 (assuming neutral stability conditions) according to U ln[(10. 0 Z )/Z ] 10Å D O . (2) Uz ln[(Z 0 ZDO)/Z ]

Values for ZO were based on categories published in Panofsky and Dutton (1984), Wieringa (1992), and Simiu and Scanlan (1986). When exposure is in¯u- enced by a uniform distribution of objects (e.g., houses, trees) that are large relative to the height of the anemometer, the zero-plane displacement height (ZD )is used to estimate the effective height (Z 0 ZD ) of the anemometer above the surrounding roughness elements [ZD is approximately 75% of the mean height of the objects (Arya 1988)]. It should be kept in mind that roughness estimation is very much an art; reasonable estimates of roughness by experienced investigators can vary by as much as 0.25±0.50 m. For a mean wind speed of 35 m s 01 at 20 m in neutral atmospheric stability with ZD Å 0.0, this difference in estimates of ZO 01 FIG. 5. Flow chart of procedures necessary to achieve a common would correspond to 10-m-level winds of 29.5 m s 01 framework for analysis of hurricane wind observations. (ZO Å 0.25 m) and 28.4 m s (ZO Å 0.50 m), respectively. Similarly, an adjustment of a 35 m s 01 wind from 5 m would result in 10-m-level winds of 43.1 and 45.5 m s 01 , respectively. These calculations imply a the trajectory. In a hurricane the ¯ow trajectory is com- velocity uncertainty of {7% for a 0.25-m difference in plicated by additional accelerations associated with the roughness length estimation. In contrast, based on these pressure ®eld, the radial distance from the storm center, examples, failure to adjust winds to a common refer- and convective components associated with updrafts ence height could lead to errors in wind estimates at a and downdrafts in the eyewall and rainbands. given level of 15%±30%. This error becomes much larger as the difference between actual anemometer b. Adjustment of wind measurements to 10 m height and the standard reference height increases. Surface winds in hurricanes can vary greatly hori- 2) MARINE EXPOSURE zontally depending on the roughness of the upwind terrain and the distribution and size of obstacles along the Despite recent studies of drag, heat, and moisture ¯ow. This variability was readily apparent in ground coef®cient dependence on wind speed, most notably and aerial observations of the damage ®eld after An- the Humidity Exchange Over the Sea experiment (Kat- drew. Structures exposed to ¯ow over open terrain were saros et al. 1987), few observations are available to much more susceptible to damage than those sheltered test the suitability of these relationships in hurricane by surrounding houses or trees upwind. Although most force wind speeds. We use the same model (Liu et al. NWS and FAA weather stations and forecast of®ces 1979) employed operationally by the National Oceanic prescribe open exposures for installation of wind mea- and Atmospheric Administration (NOAA) National suring equipment, not all anemometers are in compli- Data Buoy Center for the reduction of Coastal Marine ance, and much variability exists in the private sector. Automated Network (C-MAN) and buoy mean winds For meteorological applications the height recom- to the 10-m level. The model is also used to calculate mended by the WMO for wind measurement is 10 m the roughness length for determining surface stress above ground level. To properly adjust winds to a com- from lowest-level eta model winds over the ocean (Jan- mon height, the upwind fetch roughness must be jic 1994). This model includes interfacial sublayers known. Similarity relationships form the basis for close to the sea surface where molecular effects on height correction of wind speed measurements dis- transports are important. Surface stress and roughness

/3q05 0218 Mp 311 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 312 WEATHER AND FORECASTING VOLUME 11

FIG. 6. LLWAS data in the vicinity of Miami International Airport for 0934 UTC during Hurricane Andrews landfall. (a) Roughness lengths (ZO), anemometer heights (ZAN), zero-plane displacement heights (ZD), and wind barb plots of the 1-min mean wind observations at 0934 UTC. (b) 10-min mean wind observations centered on 0934, adjusted to 10 m and open terrain. Wind barb convention is 25 m s01 for a triangle, 5 m s01 for a single barb, and 2.5 m s01 for a half barb. are computed according to stability as determined by smoother roughness values over water relative to those the air±sea temperature difference. For hurricane over land, errors in roughness length parameterization winds, the model was con®gured to use the drag co- are smaller than those estimated over land. For a ef®cient dependence on wind speed developed by {0.002-m variation in roughness length about a value Large and Pond (1981). of 0.01 m, the wind speed difference is about {3%. As discussed by Liu et al. (1979), this model's treat- Failure to adjust marine wind speeds to 10 m will result ment of interfacial sublayers may be suspect at high in errors comparable to those previously mentioned for wind speeds, especially when spray effects alter tem- land exposures (i.e., 15%±30%). perature and humidity pro®les in the surface layer. The effect of ocean wave age on sea surface stress (Janssen 3) ADJUSTMENT OF FLIGHT-LEVEL WINDS 1989) may also be important; in a fast-moving hurricane like Andrew, where the wave ®eld at any point is During Andrew's landfall, air force reconnaissance subject to changing wind speed and directions, the sea aircraft recorded data at a ¯ight-level of 2.5±3.2 km surface may be rougher than a fully developed sea. and transmitted the observations in real time to NHC Hence, the changing wave ®eld could act to further via an aircraft±satellite data link. For the ¯ights on increase uncertainty in the roughness. Florida's southeast coast, the wind data consist of con- 01 For 10-min mean winds of 30 and 60 m s at ZA secutive 1-min means and the peak 10-s mean within Å 39 m at Fowey Rocks C-MAN station, the Liu et al. each minute; complete 10-s data were not available. For (1979) model computed ZO at 0.007 and 0.015 m, re- the ¯ights on Florida's southwest coast, complete con- spectively. The ZO value of 0.015 m is close in mag- secutive 10-s and 1-min mean wind data were avail- nitude to values of ZO Å 0.03 m normally attributed to able. The data were collected along radial ¯ight tracks open-terrain land roughness. Based on these roughness extending from the eye to over 220 km outward. Po- calculations, for winds on the order of 60 m s 01 , one sitions and wind measurements are determined via the should expect little change in mean wind speeds when Self-Contained Navigation System (SCNS), based on ¯ow progresses from the ocean to land with open- solutions determined from an inertial navigation unit. terrain characteristics. Because of characteristically Position accuracy is advertised as {4.3 km; actual ac-

/3q05 0218 Mp 312 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 313

FIG. 7. Examples of contrasting exposures for LLWAS sites FA1 (a) and FA3 (b) plotted in Fig. 6. curacy is probably much better since typical terminal able near the time of landfall). For the ¯ight covering errors are less than {2 km (R. Katz, SCNS program Andrew's emergence over the Gulf of Mexico, com- manager, 1993, personal communication). Reconnais- plete 10-s data were available. Time series of these data sance aircraft attempt to meet or exceed accuracy re- were examined for obvious spikes and dropouts, which quirements of {2ms01 for ¯ight-level wind speed and were edited from the dataset. {5Њ for wind direction (OFCM 1993b). No intercom- Flight-level observations are adjusted to the surface parison ¯ight data were available for Andrew or any with a planetary boundary layer (PBL) model (Powell other hurricanes during the 1992 season. An intercom- 1980). Flight-level winds are assumed to be represen- parison of NOAA and air force aircraft conducted in tative of mean winds in the boundary layer and Dear- Hurricane Emily of 1993 and analyzed by the authors dorff 's (1972) de®cit law approach is followed to ad- discovered occasional dropouts or negative spikes in just this wind to the top of the surface layer. The model the air force 10-s mean data on outbound southward iterates on a friction velocity solution, and then a 10- and inbound westward ¯ight legs (see Black and Hous- m-level wind is computed according to similarity the- ton 1994 for more details). These problems were ory using a roughness of 0.03 m over land and a rough- caused by errors in the aircraft data processing software ness over water based on the Charnock (1955) rela- that have since been corrected. Accordingly, for the tionship with constant a Å 0.0144 (Garratt 1977). The southeast coast landfall analysis, one outbound south- 10-m wind computed by the model is assumed to be ward ¯ight leg was removed (two northward inbound equivalent to a 10-min mean measured by a marine legs covered the same area) and only peak 10-s ¯ight- surface station; this quantity is then adjusted (see be- level data were considered for the inbound westward low) to a maximum 1-min mean that would be ex- ¯ight leg (where no additional ¯ight legs were avail- pected within a 10-min period. Thermodynamic input

/3q05 0218 Mp 313 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 314 WEATHER AND FORECASTING VOLUME 11 to the model assumed neutral conditions over land in except for eyewitness reports of calm and the onset of the vicinity of the eyewall (surface temperature 0.1ЊC strong southerly winds on the back side of the eye, no cooler than the air temperature), slightly stable con- surface observations were available to validate the as- ditions (surface temperature 1.0ЊC less than air tem- signment of wind directions within the eye; this will be perature) radially outward from the core over land (ra- discussed further in Powell and Houston (1996). We dii greater than 34 km), and unstable conditions further note that asymmetric in¯ows (Powell 1982) throughout over water (sea surface temperature 3.5ЊC and even out¯ows (Powell 1987) have been observed warmer than the air) as observed by Fowey Rocks C- at the surface in landfalling hurricanes related to a MAN station. Adjustment ratios from this model were background environmental ¯ow and outer convective inversely proportional to the magnitude of the bulk rainband development, respectively. No such asym- 2 Richardson number [de®ned as gZBDTA-S /(TAV )], metries were apparent in the Andrew observations. where g is gravitational acceleration, ZB is the boundary Obviously, the aircraft was too high to be considered layer height (assumed 750 m), DTA-S is the air±sea near the PBL. Multiple ¯ight-level research experi- virtual potential temperature difference, TA is the vir- ments and airborne Doppler radar measurements indi- tual potential temperature, and V is the ¯ight-level wind cate that maximum mean winds in hurricanes are usu- speed) and ranged from 60%±80% over land to 70%± ally found at 0.5±2.0 km, below the 2.5±3.2-km level 90% over water. Higher wind speeds reduced the bulk ¯own in Hurricane Andrew. However, a comparison Richardson number to more neutral values, causing (Powell and Black 1990) of 69 surface measurements lower adjustment percentages. in unstable conditions (sea surface temperature greater Past studies (e.g., Graham and Hudson 1960) and than buoy air temperature) to aircraft wind observa- more recent data collected in hurricanes simultaneously tions from 500-m, 1500-m, and 3000-m levels indi- monitored by reconnaissance aircraft and NOAA buoys cated no consistent difference in the ratio of surface to or platforms (e.g., Powell 1982; Burpee et al. 1994) ¯ight-level winds. In addition, convective updrafts and indicate that surface in¯ow is maintained in the eyewall downdrafts in eyewalls of mature hurricanes tend to and within the eye and that differences between ¯ight- vertically mix peak winds above the boundary layer, level and marine surface observations are on the order which would act to decrease the difference between of 20Њ. Comparisons to surface observations over land 700-mb winds and those at lower levels (e.g., Jorgen- in Andrew (see Tables A2 and A3) suggest that a rea- sen 1984; Franklin et al. 1993). While the peak winds sonable estimate of the surface wind direction over land may have been higher at lower levels, a mean PBL may be made by subtracting 40Њ from aircraft measured wind would also have to include winds affected by fric- wind directions. For this study, marine and land surface tion at lower levels, so it is conceivable that a PBL- wind directions for the adjusted aircraft observations mean wind might be of similar magnitude to the 2.5± were assigned by simply subtracting 20Њ and 40Њ, re- 3.2-km-level wind. Lacking information on what the spectively, from the ¯ight-level wind directions. The winds may have been at levels below 3 km in Andrew, resulting wind directions maintain in¯ow toward the we assume that the wind speeds at 2.5±3.2 km are rep- circulation center within the eye. We should note that, resentative of mean winds in the PBL.

FIG. 8. Schematic of wind pro®le change and internal boundary layer (height hI) development during ¯ow from marine to land exposure. Height hE represents the depth of the layer in equilibrium with the new underlying surface.

/3q05 0218 Mp 314 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 315

In order to account for a possible underestimate of rience much greater wind speeds than an adjacent area the eyewall winds by the 1-min data, maximum 10-s with rougher terrain; an observation that became ob- mean ¯ight-level wind speeds were input to the PBL vious in examining the hurricane damage. Because of model within 34 km of the center. The maximum 10-s uncertainties in estimation of local roughness and the mean winds accounted for the possibility of enhanced in¯uence of larger-scale upwind fetch, it was important transfer of momentum in highly convective regions of to test roughness estimates by comparing actual open- the eyewall. Outside the core (greater than 34 km from exposure wind measurements to neighboring rough-ex- the center) 1-min ¯ight-level wind speeds were input posure observations adjusted to open terrain. The form to the model except for the westward inbound ¯ight leg of (3) is relatively insensitive to ZO estimation errors on the east side of the storm (as discussed above for mentioned in a previous section. Applying the log law the southeast coast landfall). Maximum 10-s mean to a range of roughness lengths {0.25 m about a winds were used for that entire leg. ``true'' value of 0.5 m, and a 20±50 m s 01 range of The surface-layer formulation in the PBL model is winds, the computed 10-m-level wind can vary {14%± subject to the same limitations for high-wind drag co- 23% from the true wind speed. However, after then ef®cients and wave effects as the Liu et al. (1979) applying (3) to convert the 10-m winds to open terrain, model. Based on comparisons in a storm-relative co- the speeds vary by only {5% from the true value. ordinate framework (appendix A), boundary layer Hence, after adjusting a wind measurement to 10 m and model adjustments of ¯ight-level winds to maximum then to open terrain, the cumulative error caused by 1-min winds at 10 m in open exposure yield speeds uncertainty in estimating roughness length is on the within 20% of the actual surface winds with directions order of {5%. Failure to adjust wind speeds to standard that are within {20Њ. Failure to adjust ¯ight-level winds terrain could result in U10 values that are about 40% to the surface will produce 20%±40% overestimates of too low and prevent an analysis from resolving mean- the maximum surface wind speed and wind directions ingful differences. that have little in¯ow. In general, wind measurements from marine exposures were not converted to a standard terrain but are c. Adjustment to standard (open) exposure representative of the existing sea state, subject to limitations mentioned above. However, in some cases, Wind standards published in building codes (design marine observations may be available in a location winds) prescribe an open exposure typical of most air- relative to the storm where no land measurements sur- port weather station anemometers. Since knowledge of vived; such was the case in Andrew for C-MAN ob- the surface winds that occurred during Andrew may servations from Fowey Rocks. Consecutive 10-min impact future design wind criteria, we converted all mean and hourly 2-min mean winds measured at input data over land to an open exposure based on pho- Fowey Rocks ceased at 0800 UTC when Andrew's tographic roughness characterizations of all major wind wind center was 20 km away, approximately the same measurement locations. Roughness characterizations radial distance from the storm as the maximum wind were made according to descriptive categories dis- measured by the air force reconnaissance aircraft at cussed above. All surface wind measurements over 0810 UTC. Since the Fowey Rocks observations were land were converted to a standard terrain roughness made in the vicinity of the eyewall in the northwest length of 0.03 m. By applying the relationship between quadrant of the storm (where few observations were friction velocity and the gradient wind (Blackadar and available over land), we converted the adjusted 10- Tennekes 1968; Gill 1968) to two different roughness m-level winds to what would be experienced over land categories with the same gradient wind, Simiu and in open terrain using (3) with u˙ and ZO supplied by Scanlan (1986) determined the Liu et al. (1979) model and then applied the log law for neutral conditions. u˙ Z 0.0706 SOSÅ . (3) u˙ ͩͪZO d. Adjustment to maximum sustained wind speed

In this case, u˙S is the friction velocity appropriate for Since the maximum sustained (1-min average) wind the standard terrain roughness, u˙ is the friction veloc- VM1 is released in advisories and warnings issued to the ity appropriate for the wind measurement adjusted to public by NHC, all available wind data were converted 10 m as determined from the log law, ZOS is the stan- to this quantity if possible. After adjustment to 10 m dard open-terrain roughness of 0.03 m, and ZO is the (and adjustment to open terrain for land exposures), roughness length consistent with the exposure for the observations with averaging times greater than 2 min given wind measurement. If the actual U10 for a rough- (including adjusted aircraft observations assumed to be ness of 0.5 m was 26 m s 01 (u˙ of 3.5 m s 01 ),the(3)'s equivalent to 10-min mean surface winds) were ad- 01 solution of u˙S is 2.85 m s , and the open-terrain neu- justed to the probable maximum 1-min wind within the tral stability U10 (computed from the log law) is 41.5 observation sampling period. The adjustment proce- ms01. Hence, an area with open terrain would expe- dure used relationships between the 1-min gust factor

/3q05 0218 Mp 315 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 316 WEATHER AND FORECASTING VOLUME 11

and sampling period, based on the work of Durst able for determining s60,T , and without complete wind (1960) for level terrain ¯ow, Krayer and Marshall records it is not known whether the stationarity as- (1992, hereafter KM) for hurricane winds over land, sumption was satis®ed for those data. 3) The data avail- and additional hurricane data collected from NOAA able to create (6) include both land (open terrain) and buoys since 1979. This work assumes that deviations marine exposures; as more high-resolution digital wind of short-period wind averages from the longer-period velocity records become available in tropical cyclones, mean over stationary time periods of at least 1 h follow separate G60,T relationships will be needed for marine a Gaussian distribution. Unfortunately, no high-reso- and open-terrain exposures. This technique cannot be lution wind speed data were available from Hurricane applied to typical airways hourly observations (1-min Andrew, but observations collected in Hurricane Bob average) in the Hurricane Andrew dataset because the (Fig. 4b) support this assumption. Krayer and Marshall wind was not monitored continuously over each hour. de®ne the gust factor Gt,T1 as the highest t-period mean For 5-, 10-, and 15-min average wind observations that occurring over a longer averaging time period (T1): compose the balance of the dataset, the probability of a maximum 1-min wind greater than that computed us- ∗ Gt,T1 Å 1.0 / st,T1 SND. (4) ing (6) is 20%, 10%, and 7%, respectively.

In (4), st,T1 represents the standard deviation of the short t-period mean wind about the longer T1 period 5. Quality control mean normalized by the longer-period mean wind and SND is the standard normal deviate associated with a Many reports obtained after landfalling hurricanes require quali®cation due to uncertainties in position, probability of 1 0 t/T available from probability ta- 1 time, exposure, or measurement. Evaluations of surface bles of the normal distribution (e.g., Box et al. 1978). and ¯ight-level observations were performed by near- A G relationship was developed for several aver- 60,T est neighbor comparisons of plotted information in aging periods between 120 and 3600 s. Following storm-relative coordinates (e.g., Fujita 1963; Miller KM's approach, measurements of s from the pub- t,T 1964). This technique was used in reconstructing wind lished sources were used with ®elds of Hurricanes Frederic, Alicia, Hugo, and Trop- 222 st,T1 Åst,T2/s T2,T1 (5) ical Storm Marco (Powell 1982, 1987; Powell et al. 1991; Houston and Powell 1994). Since the wind ®eld to compute equivalent normalized standard deviations is assumed to be representative of the chosen time infor time periods in which data were not available. For terval, time periods with minimal intensity changes example, the mean gust factor G5,510 from comparisons should be chosen if possible. Plots of ¯ight-level winds of NOAA buoy measurements to aircraft observations in the eye and public reports of calms and their duration in hurricanes from 1979±85 (Powell and Black 1990) in the vicinity of Homestead were used to construct a was used in (4) to solve for s5,510 . This value and a track of the circulation center. A time period of 0445± Durst (1960) value of s5,3600 were used in (5) to solve 1130 UTC was chosen to depict the wind ®eld during for s510,3600 . Finally, (5) was solved for s60,510 given Andrew's landfall in south Florida; shorter time periods s510,3600 and a published value (Durst 1960) for s60,3600 . excluded many observations and resulted in large data- The resulting G60,T values (Fig. 9) were ®t by a third- sparse areas. From 0445 to 0915 UTC central sea level order polynomial valid for 120 s õ T õ 3600 s: pressure decreased from 93.6 to 92.2 kPa and then increased to near 94.4 kPa by 1130 UTC, while the max- G60,TÅ 2.6631 0 2.1244 log 10 T imum 1-min ¯ight-level wind speeds on the north side 01 /0.85245(log T)230 0.10346(log T) . (6) of the storm increased from 68 m s to a maximum of 10 10 82 m s 01 at 0810 UTC and then decreased to 70 m s 01 For example, this relationship increases a 10-min av- by 1010 UTC. erage wind speed by 11% (G60,600 Å 1.11) to estimate the highest 1-min average that would occur over the a. Treatment of peak wind observations 10-min period. Based on differences between KM and Durst's Gt,3600 curves published in KM and the standard Several observations of the peak wind measured dur- deviation of G60,600 from Hurricane Bob data, the ac- ing the storm or the last measurement before loss of curacy of G60,T is estimated at {0.03, resulting in a electrical power or failure of the anemometer system wind speed uncertainty of {6%. were brought to our attention after a public appeal for Accuracy of these methods depends on the relevance data was published in the Miami Herald (Wallace of the underlying Gaussian assumption. Shortcomings 1992). Frequently, there were no written or otherwise include the following. 1) It is not possible to convert recorded values, and details of the observations were any 1-h averaging time of say, 1 min, to another of 10 determined from interviews and site visits. Most cre- min; we can only estimate the highest short-period dence was given to sites with well-exposed anemom- wind that might be observed over some longer time eters; several homeowner anemometers were placed period, or vice versa. 2) Very few datasets were avail- too close to the roo¯ine to provide an accurate wind

/3q05 0218 Mp 316 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 317

FIG. 9. Dependence of 1-min gust factor on averaging period based on data from Krayer and Marshall (1992), Durst (1960), and NOAA marine platforms in hurricanes. measurement. Interviews or written reports obtained on nearest-neighbor comparisons of storm-relative po- within 1±2 months of the storm were considered to be sitions of the observation for several possible times most accurate; followup interviews sometimes led to (PER 0830, 0845, and 0900 UTC) and the location of statements that contradicted earlier interviews. When the observation relative to the eyewall (dashed line) in an observation based on an interview or log was in Fig. 10. In this case, PER 0900 UTC was located on doubt, plotting it relative to surrounding observations the inner side of the eyewall and ®t in best with the of known quality made it possible to accept or remove 0759 UTC observation from Fowey Rocks C-MAN the datum. (marine exposure) and the peak adjusted winds from Only one visual observation without a log was the aircraft (which are farther inward than the ¯ight- deemed suf®ciently accurate to be included in the An- level winds depicted in Fig. 10). It is possible that drew surface wind observations. This report from Per- slightly higher winds may have occurred at this site rine (PER) was a visual record of the last digital read- following the instrument failure, but the PER 0900 ing of the instrument (95 m s 01 ) before the 10-m mast UTC value is consistent with the highest surface wind failed. The instrument did not survive, but we acquired estimates computed from the aircraft data. three identical samples from the manufacturer for wind Of special concern were reports of anemometers tunnel testing (see appendix B). Tests indicated that ``pegged'' for periods of several minutes at Tamiami the instruments over estimated the true wind speed by Airport and at north Key Largo. Examination of ar- 16.5%, suggesting that the actual gust was 79.4 m s 01 . chived anemometer charts from severe hurricanes (in- A recent study by Black (1993) suggests that gust fac- cluding Camille of 1969, Celia of 1970, and Frederic tors in landfalling hurricanes decrease with height at of 1979) con®rms that eyewall winds are extremely approximately the same rate that the mean wind in the turbulent, especially over land, with high-frequency boundary layer decreases with height, implying that gusts and lulls. An envelope of the lulls is typically peak convective downdraft gusts conserve momentum about 40%±60% of the gust speeds. Given this ratio, a vertically. Hence, the Perrine gust was assumed to be ``pegged'' value of 50 m s 01 at Tamiami Airport would associated with a convective downdraft in the eyewall imply peak gusts of 100±125 m s 01 and sustained and was not adjusted for exposure. A gust factor curve winds of between 77 and 100 m s 01 . This was consid- published in KM was used to convert the gust (assumed ered extremely unlikely since the last measurement of to be a 2-s mean) to a hypothetical 1-h mean (79.4/1.7 sustained winds at Fowey Rocks C-MAN station (with Å 46.7 m s 01 ), and the maximum 1-min wind for the marine exposure and closer to the radius of maximum hour was estimated as [(46.7)(1.32)] 61.6 m s 01 . This winds) was 55 m s 01 at 10 m. It is possible that the observation occurred ``sometime'' between 0830 and wind speed circuit of the F420 anemometer (which was 0900 UTC, and no wind direction was given. Time and in the process of falling off the crossarm before the wind direction of the observation were estimated based tower failed, Fig. 2) picked up voltages from the di-

/3q05 0218 Mp 317 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 318 WEATHER AND FORECASTING VOLUME 11

01 rection circuit causing the speed dial to read offscale VuuÌV (R. Marshall, National Institute of Standards and Tech- tanf Å , (7) ͩͪͩͪRMW Ìz nology, 1993, personal communication). Of®cial reports from Tamiami Airport through 0833 UTC were where Vu is the tangential wind component, z is height, considered reliable. However, the peak Tamiami Air- and ÌVu /Ìz is the vertical shear of the tangential wind. port observation attributed to 0848 UTC was elimi- According to the thermal wind relationship given in nated from this dataset although it is considered valid Jorgensen (1984), in other descriptions of Hurricane Andrew's winds 01 (e.g., May®eld et al. 1994). 2Vu 01 ÌT ÌVu/Ìz Å f / fg(fTm) , (8) Additional pegged anemometer reports were con- ͫͩRMW ͪ ͬ ͩÌr ͪ P tained in the log of the sailboat Mara Cu, moored north of Key Largo (Fig. 10). According to the manufac- where f is the Coriolis parameter, g is the gravitational turer, the measurement display indicated a 2±5-s mean acceleration, Tm is the mean temperature of the layer ap- wind, and the wind direction was given relative to the plying to the shear calculation, and (ÌT/Ìr)P is the radial heading of the vessel. The display was limited to two gradient of temperature along a constant pressure surface. digits in knots for wind speed, and the instrument was According to these relationships, an intense, compact hur- 01 01 rated as accurate to {2ms up to 26 m s (V. ricane with a small RMW will display very little outward Valldeperis, Brooks and Gatehouse, 1995, personal slope of angular momentum surfaces. Indeed, Hawkins communication). The deck log indicated that the read- and Imbembo (1976) found a near-vertical RMW in Hur- ing was in excess of 51 m s 01 at 0833, 0840, and 0846 ricane Inez (1966), a very small intense hurricane very 01 UTC, and the captain of the vessel estimated the winds similar in size (13 km RMW ) and strength (67 m s mean to be much higher based on the noise. Flight-level winds at 500 m with 10-s gusts to 81 m s01 ) to Andrew. winds adjusted to the surface in the vicinity were Ç50 This calculation (Table 2) was made for Andrew based ms01, implying gusts of Ç65 m s 01 . Given that lulls on measurements made on the 70.0-kPa pressure surface are typically 40%±60% of gusts, the Mara Cu winds ¯own by the air force aircraft. According to Table 2, RMW are too strong to be considered lulls; it is likely that tilt increased from 50Њ±60Њ to near 80Њ, while Andrew lulls occurred during intervals between the log entries. intensi®ed from 0600 to 0900 UTC and then tended to Given the uncertainty of what the gust and lull values decrease during landfall from 75Њ to 64Њ. On the south may have been during these periods, and the fact that side, the RMW was initially steep (due to relatively small nearby surface-adjusted ¯ight-level winds were of the RMW and high VT )at69Њand then decreased (because of same order as the log records, we assumed that the an increase in RMW and a decrease in VT ) to a more out- winds recorded in the Mara Cu log were representative ward slope of 45Њ during landfall. of 5-min means. These measurements are unique because safety con- siderations have previously restricted aircraft from following a hurricane over land during landfall. The cal- b. Position of the radius of maximum winds culations suggest that the landfall process may cause the RMW to tilt more outward as the winds decrease and To properly adjust ¯ight-level winds to the sur- that the effect may be asymmetric. Estimates of the face, it is necessary to know the location of the radius RMW at the surface, assuming the tilt calculated in Table of maximum sustained winds (RMW ) and its variation 2, suggest a surface RMW 1±3 km inward of that deter- with 1) height from the surface to the altitude of the mined from the 1-min mean ¯ight-level winds. Radi- air force reconnaissance aircraft (3 km), 2) azimuth ally outward from the wind maxima, the time of the relative to the heading of the storm, and 3) intensity peak 10-s wind within each 1-min ¯ight-level obser- as the storm weakened during landfall. Failure to ac- vation in the storm core was simply assigned to coin- count for tilt in the vicinity of the eyewall places the cide with the inward most possible 10-s period without adjusted maximum winds from ¯ight level too far regard to the computed surface RMW location. For ex- from the surface center of the storm. Unfortunately, ample, on the south side of the storm, the time of the no continuous anemometer records were available in peak 10-s value measured by the northbound aircraft the core region to indicate the position of the surface between 0801:00 and 0802:00 UTC was taken to be RMW . Studies by Hawkins and Rubsam (1968), Jor- centered at 0801:55 UTC. The peak 1-min ¯ight-level gensen (1984), and Willoughby (1988, 1990) note winds measured in Andrew occurred on a northbound that above the boundary layer, in the azimuthal-mean leg in the northern eyewall and reached 82 m s 01 , with sense, hurricane winds are in approximate gradient a peak 10-s mean of 83.6 m s 01 at 0810 UTC (the and thermal wind balance. The tilt (f) from the hor- minute corresponding to 0810±0811 UTC). The peak izontal of the tangential wind maximum in Hurricane 10-s value was attributed to 0810:05 UTC and became Allen (1980) was shown by Jorgensen (1984) to be 60.4 m s 01 when adjusted to 10-m open exposure over very close to the tilt of lines of constant angular mo- land. Hence, PBL model-adjusted aircraft winds in the mentum computed according to core region (vicinity of the eyewall less than 34 km

/3q05 0218 Mp 318 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 319

FIG. 10. Storm-relative (for 0900 UTC 24 August 1992) locations of ¯ight-level wind speeds relative to the location of peak storm surge values measured at Burger King Headquarters (BKHQ) and the Deering Estate County Park (DEER), peak gust measured at the surface in Perrine (PER, 0900), Fowey Rocks C-MAN station (marine-exposure 10-m winds), and extreme structural damage at Naranja Lakes (NAR, 0845). Also shown are locations of calm in the eye; the minimum surface pressure measurement; the onset of strong south winds at the surface in the eyewall based on eyewitness observations at Homestead (HSD), Florida City (FLC), and Leisure City (LES); and maximum surface winds in the southern eyewall (Mara Cu). Dashed circle represents a trace through the peak re¯ectivity of the eyewall according to the Melbourne WSR-88D radar at 0902 UTC. For clarity, several aircraft and Mara Cu observations were omitted. from the storm center) displayed sharper peaks that the time of severe damage to a townhouse relative to were shifted closer to the circulation center than the 1- the locations of the maximum 1-min ¯ight level winds. min data alone would allow. The 10-s adjusted RMW are The maximum recorded storm surges (May®eld et al. shown in Table 2 as well as RMW estimated from the 1994) were 5.15 m above the National Geodetic Ver- thermal wind shear. In general, the assigned positions tical Datum at Burger King Corporation (BKHQ at are consistent with the thermal-wind-shear-estimated 15.6-km radius) and 5.06 m at the Deering Estate Park positions except for the east pass and the second south (DEER at 17.4-km radius). The surface-adjusted RMW pass, where the peak 10-s wind occurred inward from from the maximum 10-s aircraft winds (Table 2) com- the time of the peak 1-min wind. Outside the core re- pare very well with the positions of maximum storm gion (greater than 34 km from the storm center), 1- surge. The time of damage to the Naranja Lakes de- min PBL model-adjusted aircraft winds were not velopment was documented by the time read off a shifted in position. stopped clock found in the building debris (P. Black, Indirect measures of the surface RMW were made by HRD, 1992, personal communication). Three of the noting the location of peak storm surge (Fig. 10) and seven deaths within structures during Andrew's south

/3q05 0218 Mp 319 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 320 WEATHER AND FORECASTING VOLUME 11

Florida landfall were in this development despite its sides of the storm, perhaps in response to the displace- location in a storm surge evacuation area. The clock ment of the pressure center. pinpointed the time of damage, plotted as NAR 0845 in Fig. 10. The western eyewall was over Naranja c. Final wind observation distribution Lakes at 0845 UTC. The last observation from Fowey After completion of the standardization steps and Rocks C-MAN station before its mast failed was co- quality control described above, the distribution of incident with the eyewall (21-km radial distance), inland-exposure observation locations over the Ç7-h dicating that failure occurred very close to the position time period chosen to apply to southeast Florida is dis- of the R . MW played in Fig. 11 with geography corresponding to the Several observations of the eye (Fig. 10) were sup- time of landfall (0900 UTC). Note the change in data plied by the public (sources are listed in the acknowl- density from the west to the east side of the storm center edgments). Calm winds in the eye were observed in associated with anemometer system failures. A sepa- Homestead (HSD, 0929) and at Florida City (FLC, rate distribution (not shown) contains the marine-ex- 0935). Onset of strong south winds is indicated at FLC posed observations over the same time period. These 0948±1013, Leisure City (LES, 0945), and HSD data at each site conform to the same framework for (0945). Eyewitness reports of extreme winds on the height, exposure, and averaging time and are now suit- back side of the eye based on the sound of the wind able for input to an objective analysis scheme. Analysis were treated cautiously. The noise level increases rap- of these data and observations from Andrew's exit from idly when new winds on the back side of the eye be- Florida's west coast are discussed in the companion come strong enough to carry debris. Small debris that paper (Powell and Houston 1996). had been deposited from the front side of the storm suddenly becomes airborne, and the density of the de- 6. Conclusions bris ®eld (and its attendant noise) is greater than on the front side of the eyewall, leading to the perception of A review of spectral analyses conducted in past hur- much stronger winds. Unfortunately, no surface mea- ricanes suggest that the turbulent, convective, and per- surements are available to document wind conditions haps mesogamma scales contribute most to the energy in the Homestead area during that time. The minimum of the wind ¯uctuations. Surface-layer similarity theory pressure in the eye (May®eld et al. 1994) was mea- scaling is applicable to horizontally homogeneous mo- sured northwest of the circulation center, from 0900 to tion characterized by mechanical and buoyancy-pro- 0915 UTC about midway between the circulation cen- duced turbulence in quasi-stationary conditions. Simi- ter and the eyewall. Calm winds were not observed at larity methods advocated here do not take into account the location of the pressure minimum. A separation of possible scaling factors associated with curved ¯ow, pressure, wind, and radar centers is probably quite upstream roughness changes, clouds, and mesoscale common in hurricanes but dif®cult to document due to phenomena but appear reasonable for height adjust- lack of data. Relative to the direction of motion, An- ment of Ç10-min mean winds in the surface layer. drew's wind center was displaced to the left of the radar Maximum 1-min winds were computed from longer- center and to the left and behind the pressure center. period averages (or in the case of Perrine from a single When compared relative to the positions of maximum peak gust value) by using gust factor relationships that winds in the southern eyewall from the sailing vessel assume a Gaussian distribution of normalized devia- Mara Cu and the positions of the maximum ¯ight-level tions of short-period means about longer-period means. aircraft winds on the east and south sides, it appears This assumption is supported by high-resolution sur- that the RMW was only 10±15 km on the south and east face data collected during Hurricane Bob (1991).

TABLE 2. Calculations of the tilt of the RMW in the eyewall of Hurricane Andrew. Here, VT is peak (1-min average ¯ight-level wind speed); dVZ thermal wind shear; Tilt slope of the eyewall wind maximum from horizontal; 1 min RMW (RMW based on maximum 1 min ¯ight-level winds); SFCRMW RMW at the surface computed from tilt; and 10-sRMW RMW based on estimated position of peak 10-s wind at ¯ight level.

Eyewall Time of pass 1 min RMW VT dVZ Tilt SFCRMW 10-sRMW location (UTC) (km) (m s01) (m s01 km01) (deg) (km) (km)

North 0806±0813 19.6 82.0 00.95 77.2 18.9 18.7 0915±0920 20.9 76.8 01.08 73.6 20.0 17.8 1012±1019 18.0 70.1 01.51 68.9 16.8 15.1 South 0757±0803 15.9 61.4 01.27 71.8 15.0 13.6 0921±0926 22.2 50.5 02.29 44.8 19.2 11.6 1002±1010 21.9 53.1 02.24 47.2 19.1 19.5 East, west 0544±0553 23.0 62.9 02.00 53.7 20.8 15.9 0553±0558 19.8 66.4 01.74 62.7 18.2 17.8

/3q05 0218 Mp 320 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 321

FIG. 11. Final distribution of wind data locations for Andrew's landfall in southeast Florida. Geography is positioned relative to Andrew's circulation center at 0900 UTC 24 August 1992. Crosshairs are plotted at 1Њ latitude±longitude intervals. Solid line represents the storm track.

In order for a wind ®eld analysis to be useful to fore- For the present, the procedures applied to the Hur- casters or the public, all input data should conform to ricane Andrew observations appear to resolve many of a common standard framework. This framework is de- the wind speed discrepancies caused by differences in ®ned by the warning and advisory requirements of terminology, exposure, measurement height, averaging NHC, guidelines for surface wind measurement, and period, and terrain and are simple enough to be used in by wind loading provisions of building codes. Here, we analyzing a historical or real-time wind ®eld. These de®ned the framework as the maximum 1-min sus- procedures are only valid for relatively uniform terrain tained wind speed at the 10-m level over open terrain typical of many coastal areas. They are not applicable (for land analyses) or over an observed or modeled sea to locations characterized as mountainous or complex state (for oceanic analyses). Hence, all input data must terrain. In such areas, a marine-exposure analysis undergo screening and adjustment procedures to con- should be conducted, and the resulting ®eld then used form as closely as possible to the framework. A mea- as an initial condition to a mesoscale model capable of sure of the error of each standardizing procedure and resolving ¯ow over complex terrain or to a wind tunnel the consequences of not following each procedure is model of the effected area. The companion paper shown in Table 3. For a 50 m s 01 wind speed, assuming (Powell and Houston 1996) examines objective the errors in Table 3 are representative of standard de- streamline and isotach analyses of the surface wind viations, the total error was computed by summing the ®eld and potential products for real-time disaster re- variance contributions of the height-exposure and time sponse, recovery, and damage mitigation activities. procedures. For oceanic observations the error is on the Anemometer system failures in landfalling hurri- order of 7% while it is 10.5% for land observations. canes will continue to occur until survivable design,

/3q05 0218 Mp 321 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 322 WEATHER AND FORECASTING VOLUME 11

TABLE 3. Estimated accuracy and consequence error ASOS will require changes to comply with these stan- of wind standardization procedures. dards in order to improve the chances that future anemometers will survive hurricanes (Powell 1993). Procedure Consequence Procedure error error Acknowledgments. This research was supported in Marine platforms part by the Coastal Hazards element of NOAA's Adjustment to 10 m {3% 15%±30% Coastal Ocean Program and Florida Power and Light Adjustment to max 1 min Corporation (FPL) Grant 93MGD03. Bret Christoe All averaging times over 2 min {6% 15%±25% and Neal Dorst of HRD helped to perform site surveys and reduce data. Joyce Berkeley of HRD and Shirley Land platforms Murillo of MAST Academy also helped with data re- Adjustment to 10 m {7% 15%±30% duction. Much useful data were acquired with assis- Adjustment to standard tance from NHC including former director Dr. Bob exposure {5% 30%±40% Sheets, Deputy Director Jerry Jarrell, Dr. Ed Rappa- Adjustment to max 1 min port, and John Parr, and from the Miami NWS Forecast All averaging times over 2 min {6% 15%±25% Of®ce staff under the leadership of Paul Hebert. Special thanks go to the public of south Florida, who responded Reconnaissance aircraft to our request for data, especially Randy Fairbank, John PBL model adjustment to 10 m, Bigelow, Peter Skipp, Debbie Iacovelli, Allan DeParle, standard exposure, max 1 Mrs. G. Hall, Mr. B. Martens. Thanks also to Dan Ad- min (based on comparisons with nearby surface ams of FPL, Janet Bravo of Metro Dade County Water observations) {20% 20%±40% and Sewer Department, Eric Swartz of the South Flor- ida Water Management District, Tom Ross of the Na- tional Climatic Data Center, Steve Gill of the National Ocean Service, Bob Panko and DeWitt Smith of Ev- installation, and exposure guidelines are followed. erglades National Park, Amy Kazmier of Big Cypress Many of these guidelines are now contained in a stan- National Preserve, David Holmes of the Dade County dard for surface wind measurement devised at a 1992 Agricultural Extension Service, David Crane of the workshop organized by the Federal Coordinator for Florida Division of Forestry, Eric Meindl, Dave Gil- Meteorology at the request of the Interdepartmental housen, and Mike Burdette of the NOAA National Data Hurricane Conference (OFCM 1992). This standard is Buoy Center, George Priest, David Fleming, Don currently undergoing submission to the American So- Groot, and Michael Greco of the FAA, and Billy Wag- ciety for Testing and Materials. With eventual adoption ner of Monroe County Civil Defense. Gale Carter and universally, a wide variety of wind equipment manu- Eric Dutton of the 53rd Air Force Reserves Weather facturers and users will have access to the information Squadron (Hurricane Hunters) and John Pavone of needed for a successful installation and archival. If the CARCAH assisted with acquisition and interpretation proposed standard is adopted in the United States, of the air force reconnaissance observations. Mike

TABLE A1. Characteristics of selected surface observation sites. Record types: d Å digital, T Å trace, I Å interview, and L Å log. Gust notes: C Å from center runway anemometer (FA1), PT Å peak within averaging time (Tavg.), and PH Å peak during hour. Za Å anemometer height, Zo Å roughness length, D Å zero-plane displacement, and Marine Å Liu model results converted to open terrain exposure with 0.03-m roughness.

Freq Station Za Record Tavg Samples Zo D ID (m) type (min) h01 Gust notes (m) (m) Source

FA2 12.2 d 160 C .30±.50 2.±3. FAA/LLWAS FA4 12.4 d 160 C .20±1.00 1.±6. FAA/LLWAS FP1 10 T 10 6 PT 0.075 0 FP&L Turkey Point FSW 60, 10 T 10 6 PT(From 60-m level) 0.075 0 FP&L 6 mi. SW of Turkey Point HAU 13.7 d 21 PH Marine Marine NOAA/NOS JOE 10.1 d 15 4 PT .01±.03 0 National Park Service MaraCu 16.7 L 52 PT .08±.30 3 J. Bigelow MBY 3.5 d 15 2 PT .01±.10 0±1.0 National Park Service MLR 14.7 d 10 6 PH Marine Marine NOAA/NDBC NHC 46 T 10 6 PT .20±1.00 3.±6. NOAA/NHC PER 10 I 3 sec 1 Final digital readout Ð Ð R. Fairbank of instrument VKY 13.7 d 11 PH 0.2 1 Metro Dade Water and Sewer

/3q05 0218 Mp 322 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 323 ,

Black, Paul Leighton, Bret Christoe, Peter Dodge, and SFC (km) DelS

Chris Samsury recorded and analyzed the WSR-57 ra- WS dar measurements in Hurricanes Andrew and Bob. Chuck Long of the U.S. Army Engineer CERC Field Research Facility provided wind data collected in Hur- 10505 1.3 0 . observation time± Time diff ricane Bob. James Acquistapace of Davis Instruments (hhmmss) FC provided the anemometers for wind tunnel testing, and S the Civil Engineering Department of the Virginia Pol- ¯ight-level wind speed and AC

ytechnic Institute and State University provided the AC wind tunnel. Much useful critical input was provided WS WD , by reviews of earlier versions of this manuscript by Gust/ anonymous reviewers and by Dr. Robert Sheets and the AC hurricane specialists of NHC, Dr. Robert Burpee, Dr. WS

Peter Black, John Kaplan, Peter Dodge, and Paul Willis SFC/AC

of HRD, Dr. Wilson Shaffer of the NWS Techniques WS Development Lab, Dr. Richard Marshall of the Na- tional Institute of Standards and Technology, Dr. John AC Peterka of the Colorado State University Civil Engi- (deg) WD neering Department, former NHC director Dr. Robert Means 0.72 0.83 Simpson, and Jeff Keppert of the Australian Bureau of ) 1 0 Meteorology Research Center. Mention of a manufac- AC WS

turer or product does not constitute an endorsement of the (m s U.S. Department of Commerce, NOAA, or the authors.

APPENDIX A ACSFC (deg)

Comparisons of Surface and Aircraft Observations WD

The accuracy of adjusting ¯ight-level measurements ratio of observed surface gust to ¯ight-level wind speed; Time diff ) 1 0 to the surface was estimated by comparison to surface AC ACSFC WS platforms in a storm-relative framework satisfying (m s WS stringent time and spatial separation criteria. Charac- teristics of the surface observation platforms used in the comparisons are shown in Table A1. Both land-

based and model-adjusted aircraft winds represent (UTC) maximum 1-min sustained winds for open exposure at A/C time

10 m as discussed above; gusts are raw values that have surface wind speed from PBL model and assigned direction for aircraft data; undergone no adjustments. Only nine independent ) to maximum 10-s ¯ight-level and surface-adjusted aircraft winds within 34 km of Andrew's wind center. Here, ACSFC (km) FCOB comparisons were possible in the core region (radial Radius S WD

distance less than 34 km) with time differences less , than 1.1 h, distance separations of less than 10 km, and ) 1 ACSFC radial and azimuthal distance separations of 5 and 10 0 51.5*51.5* 28.1 27.4 0812:05 0917:55 49.1 55.0 48 51 67.4 75.9 88 91 0.64 0.58 0.68 0.64 4755 755 2.7 2.9 WS km. Comparisons in Table A2 suggest that the sus- Gust (m s tained surface winds were 72% of the ¯ight-level peak ú ú 10-s wind speeds, with values ranging from 57% to

103%. Model-adjusted winds tended to be greater than SFC (deg)

the closest surface observations (bias of 5%) and the WD root-mean-square (rms) difference was 21% ({9 01 ) ms ). Surface-adjusted aircraft wind directions 1 0 showed an rms difference of {16Њ, and surface wind SFC WS gusts were 64%±103% (mean 83%) of the nearest (m s ¯ight-level maximum 10-s value. Much of the variability in the comparisons may be due to changes oc- ratio of observed surface wind speed to ¯ight-level wind speed; Gust/ curring over relatively small time and space differences Time (UTC) 0846:000901:00 57.4 48.8 260 231 NA NA 13.8 13.8 0802:05 1006:05 43.8 35.5 228 216 59.4 47.3 268 256 0.97 1.03 NA NA 4355 5.8 0909:000915:00 35.3 35.2 220 200 48.9 46.4 16.0 18.3 0922:55 0923:05 30.2 45.7 213 230 47.3 62.2 256 270 0.75 0.57 1.03 0.75 1355 805 5.6 2.6 between the positions of the aircraft and surface obser- SFC/AC WS vations. This variability is not surprising given the ex- A2. Comparisons of the closest surface observations ( treme radial gradient in wind speed, and the azimuthal is observed surface wind speed and direction; ABLE SFC or temporal variation in convective activity at a given FCOB T * NHC's trace went off scale. S Aircraft time; DelS Distance separation between sfc. observation and aircraft; and NA data not available. Mara Cu Mara Cu NHCPERVKY 0910:00 0900:00 0830:00 44.1 60.9 34.4 45 45 58 79.4 50.6 18.5 32.9 0809:55 0813:05 57.9 41.3 41 50 79.9 56.0 81 91 0.76 0.61 0.99 0.90 5005 1655 2.6 0.4 WD Mara Cu Mara Cu MBYNHC 0938:00 0900:00 32.4 43.1 211 45 NA 29.1 0924:05 41.3 221 55.9 261 0.58 NA 1355 8.4 radius. Also important is the comparison of the highest direction;

/3q05 0218 Mp 323 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 324 WEATHER AND FORECASTING VOLUME 11 (km) DelS 12530 2.0 12430 0.6 10430 4.0 11830 2.4 0 0 0 0 Time diff (hhmmss) AC WS Gust/ SFC/AC WS AC (deg) WD ) 1 0 AC WS (m s Means 0.63 0.78 ACSFC (deg) WD ) 1 0 ACSFC (m s WS (UTC) A/C time (km) Radius ) 1 A3. Comparisons of the closest surface observations to maximum 1-min ¯ight-level 0 51.5 28.1 1018:30 41.3 45 59.6 85 0.72 NA and surface-adjusted aircraft winds greater than 34 km from Andrew's wind center. Gust § (m s ABLE T SFC (deg) WD ) 1 0 SFC WS (m s Time (UTC) 0615:00 14.4 320 NA 83.8 0605:30 19.2 346 27.6 26 0.52 NA 00930 3.8 FCOB S FA2 0855:00 27.2 56 34.5 41.4 1020:30 33.9 52 48.8 92 0.56 0.71 FA2FA2FA4FA4 0905:00FP1 0915:00FP1 0855:00FP1 26.1 0905:00FP1 26.9 0725:00FSW 33.3 0735:00FSW 32.2 64 0745:00FSW 20.7 69 0755:00FSW 22.9 0525:00 57FSW 25.7 0545:00 64FSW 315 31.2 35.6 0555:00 12.2FSW 310 32.5 0615:00 11.1FSW 310 34.5 0625:00 11.9 40.9HAU 305 35.6 0705:00 330 14.7 41.2 22JOE 0715:00 332 15.5 35.5 26JOE 0814:30 0725:00 335 17.8 35.1 29.1JOE 0915:30 0900:00 330 18.6 32.7JOE 1019:30 56.4 17.9 0523:00 325 19.4 34.6Mara 0916:30 50.2 Cu 44.0 17 24.0 0553:00 315 38.0MBY 0601:30 37.8 18.4 117.3 0938:00 312 36.3MBY 10.2 0600:30 25.6 1023:00 315 0559:30 41.4MLR 11.6 106.9 22.9 102.7 0558:30 55 0610:30MLR 90 33.3 0508:00 22.0 25.6 57 317MLR 28.8 94.6 0908:00 24.9 25.1 27.1 0609:30 47 0608:30 311MLR 90.6 0855:00 11.8 26.9 30.9 16.0 53 220NHC 73.6 49.8 0905:00 0607:30 31.8 338 43.7 196VKY 67.4 54.8 0915:00 0606:30 16.1 20.6 17.3 336 11.3 333 61.3 52.4 0925:00 0604:30 310 21.0 13.3 330 19.1 59.6 347 0900:00 0603:30 53.6 95 234 20.8 37.9 135.6 31.8 19.2 0830:00 0602:30 97 253 20.8 32.4 353 120.9 35.9 19.5 350 39.0 87 249 43.1 0913:30 20.0 0613:30 10.2 28.9 44.5 93 0.52 23.1 243 34.4 348 21.3 0611:30 22.9 17 36.8 0.49 236 350 23.4 16 131.9 29.1 30.3 0.64 25.0 348 14 1003:30 13.2 45 NA 0.54 347 10 0925:30 27 25.8 27.7 15.9 58 0.71 NA 0.76 344 0612:30 50.9 27.7 0.59 30.3 0.74 31.6 33 0.72 30 28.2 0759:30 50.4 0.66 28.4 63 344 0.66 0.53 29.0 50.8 28 0755:30 0.60 14.8 345 50.6 55.0 30.7 30 0.69 05030 0756:30 0.47 0.47 30.7 228 27 0.73 00030 0959:30 0.75 43.7 19.2 214 15.7 27 0929:30 0.53 32.9 0.73 0.77 22.9 346 1.8 24 18.8 0.56 01130 1.4 45.6 18.5 0.63 12330 0.73 242 103 0.74 0813:30 16.4 24 40.9 0.64 13430 237 14530 1.5 25 0.92 21.3 0.63 15630 04530 243 1.8 268 0.83 37.4 0.55 44.2 207 2.2 254 0.53 0.91 2.3 02430 206 01330 22.7 0.51 0.87 2.2 3.8 26 27.1 0.73 0.88 00730 282 2.5 26.7 0.70 01830 0.4 51 1.00 277 23.8 0.59 10030 0.55 1.9 283 0.58 11130 0.72 3.4 247 0.83 12230 246 0.91 54.0 6.3 0.79 01330 6.3 0.77 05030 0.48 6.4 0.78 01830 0.52 0.88 91 02530 0.7 1.28 6.9 05730 7.1 NA 2.4 NA 0.64 10830 1.27 2.8 05930 4.4 10830 0.94 04430 2.9 00430 7.1 5.3 3.9 01630 2.6

/3q05 0218 Mp 324 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 325 model-adjusted aircraft wind (which was slightly out- APPENDIX B side the separation criteria) with the highest surface Evaluation of Digitar Weather-Master Anemometers measurement. The highest winds measured at the surface (PER) were within 0.5 m s 01 of the highest ad- Surface data from Hurricane Andrew in south Flor- justed aircraft wind on the north side of the storm and ida included a single peak reading from a Davis Instru- 5% of the nearest adjusted aircraft surface wind. The ments Digitar Weather-Master anemometer on a 10-m highest surface gust at PER was nearly identical to the mast in the Perrine area. The peak reading was 95 closest ¯ight-level 10-s gust. ms01or 212 mph as actually read from the digital read- Outside the eyewall vicinity (radial distance greater out and was widely reported in the news media. A site than 34 km), with the space separation criteria as above visit indicated that the anemometer was well exposed and time differences of less than 2 h, 31 comparisons and warranted further evaluation to assess its accuracy. in Table A3 suggest that maximum 1-min surface The 95 m s 01 speed was well beyond the instrument's winds for open terrain were 63% of the ¯ight-level 1- speci®ed operating range. min wind speeds with ratios ranging from 47% to 90%. Three anemometer samples were obtained from the Model-adjusted aircraft winds overestimated surface manufacturer and tested in the Stability Wind Tunnel observations by 14% (2.4 m s 01 ), with an rms differ- at Virginia Polytechnic Institute and State University ence of 19% ({3.7 m s 01 ). Surface-adjusted aircraft (VPISU). The stability tunnel is a low-turbulence, wind directions showed an rms difference of {20Њ, and aeronautical closed-circuit facility. Its test section is 2 surface wind gusts were 50%±100% (mean of 78%) m by 2 m, which allowed testing of the anemometers of the nearest ¯ight-level 1-min value except near the with minimal blockage. The top speed approaches 90 Molasses Reef C-MAN station, where gusts were ms01 with no obstruction in the wind tunnel. Wind nearly 30% higher than the 1-min ¯ight-level wind velocities are uniform within less than 1% throughout speeds. the middle of the test section, and speeds are measured

FIG. B1. Initial con®guration of three sample anemometers in the Virginia Polytechnic Institute and State University wind tunnel. Tunnel section (and wind ¯ow) runs right to left. Dimension from left to right Ç1m.

/3q05 0218 Mp 325 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 326 WEATHER AND FORECASTING VOLUME 11 using a certi®ed electronic Barocell Micro-manometer vided somewhat lower velocities. Both anemometers 1 connected to a standard pitot tube. Speeds measured in and 3 failed at a wind speed of 78 m s 01 . The mode of the wind tunnel at velocities greater than about 9.0 failure was the fracture of the plastic arm holding the ms01 are accurate within less than 1%. plastic cup at the hub. It is likely that the centrifugal Prior to testing, the instruments were mounted in a forces from the spinning cups exceeded the tensile ca- lower-speed wind tunnel at Clemson University and pacity of the connection at the hub. Since the anemom- operated for several hours to allow the anemometers to eters use the rotation of a magnet in the cup assembly seat themselves. At VPISU the anemometers were ini- to produce a pulse that is counted by the meter, the tially mounted as shown in Fig. B1 with one anemom- results indicate that anemometers 1 and 3 were spin- eter oriented upstream, one downstream, and one to the ning at the same rotational velocity when they failed side of the support mast. Two sets of tests were con- and that anemometer 2 was spinning at a slightly lower ducted, one with the anemometers in this orientation rate, probably due to slightly higher friction (anemom- and one with the anemometers oriented upstream or to eter 2 consistently spun at lower rates during the ®rst the side of the support. Test procedures involved in- tests as well). creasing the wind tunnel speed in increments of about The anemometers provide reasonably accurate read- 2.25 m s 01 starting at 9.0 m s 01 . Once the wind tunnel ings up to about 40 m s 01 . Above 40 m s 01 , the mea- speed had been set, digital readings from the anemom- sured velocities tend to exceed actual steady wind eters were recorded, and the wind speed was increased speeds, and the deviation increases with wind speed. to the next increment. In the ®rst test sequence, wind The facilities available did not provide for the deter- tunnel speeds were varied between 9.0 and 54 m s 01 . mination of length constants for the Digitar anemom- The anemometers were then repositioned such that an- eters or the assessment of possible over- or undershoot emometer 1 was pointed into the wind, upstream of the when the anemometers are subjected to gusty winds. support and that anemometers 2 and 3 were oriented to Conversations with Davis Instruments staff suggested the left and right sides of the support, respectively. In that anemometers in the ®eld would generally spin this con®guration, each of the anemometers were ex- more easily than new units, and they suggested the posed to ¯ow that was undisturbed by the other ane- break-in period. Based on these conversations and con- mometers. sistency of anemometer performance, ®rst-order cor- The second set of tests involved increasing the wind rections to the Perrine ®eld data were based on the re- tunnel speed to a speed where one or more of the an- sults from anemometers 1 and 3. Thus, a 16.5% emometers failed. Anemometers 1 and 3 (Fig. B2) pro- reduction was applied to the ®eld data resulting in a vided almost identical output, while anemometer 2 pro- peak gust of 79.4 m s 01 .

FIG. B2. Wind tunnel calibration results of three anemometers shown in Fig. B1. Wind speed units are shown in meters per second and miles per hour.

/3q05 0218 Mp 326 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 SEPTEMBER 1996 POWELL ET AL. 327

REFERENCES Jorgensen, D. P., 1984: Mesoscale and convective-scale characteristics of mature hurricanes. Part II: Inner core structure of Hur- Arya, S. P., 1988: Introduction to Micrometeorology. Academic ricane Allen (1980). J. Atmos. Sci., 41, 1287±1311. Press, 307 pp. Katsaros, K. B., S. D. Smith, and W. A. Oost, 1987: HEXOS: Hu- ASCE, 1994: Minimum design loads for buildings and other struc- midity Exchange Over the Sea, a program for research on water tures. ASCE Note 7-93, American Society of Civil Engineers, vapor and droplet ¯uxes from sea to air at moderate to high 134 pp. wind speeds. Bull. Amer. Meteor. Soc., 68, 466±476. Black, P. G., 1993: Evolution of maximum wind estimates in ty- Krayer, W. R., and R. D. Marshall, 1992: Gust factors applied to phoons. Proc. ICSU/WMO Int. Sym. on Tropical Cyclone Dis- hurricane winds. Bull. Amer. Meteor. Soc., 73, 613±617. asters, Beijing, China, Int. Council of Scienti®c Unions, 104± Landsea, C. W., 1993: A climatology of intense (or major) Atlantic 115. hurricanes. Mon. Wea. Rev., 121, 1703±1713. , and S. H. Houston, 1994: Comparison of NOAA and Air Force Large, W. G., and S. Pond, 1981: Open ocean momentum ¯ux mea- wind measurements in Hurricane Emily near Cape Hatteras. surements in moderate to strong winds. J. Phys. Oceanogr., 11, Minutes of the 48th Interdepartmental Hurricane Conf., Miami, 324±336. FL, Federal Coordinator for Meteorological Services and Sup- Larsen, S. E., K. Hedegaard, and I. Troen, 1982: The change of ter- porting Research, A33±A41. [Available from Of®ce of the Fed- rain roughness problem extended to mesoscale fetches. Proc. eral Coordinator for Meteorological Services and Supporting First Int. Conf. on Meteorology and Air/Sea Interaction, The Research, Suite 1500, 8455 Colesville Rd., Silver Spring, MD Hague, the Netherlands, Amer. Meteor. Soc. and Roy. Nether- 20910.] lands Meteor. Inst., 8±13. Blackadar, A. K., and H. Tennekes, 1969: Asymptotic similarity in Liu, W. T., K. B. Katsaros, and J. A. Businger, 1979: Bulk parame- the planetary boundary layer. J. Atmos. Sci., 26, 1015±1020. terizations of air±sea exchanges of heat and water vapor in- Box, G. E. P., W. G. Hunter, and J. S. Hunter, 1978: Statistics for cluding the molecular constraints at the interface. J. Atmos. Sci., Experimenters. Wiley and Sons, 653 pp. 36, 1722±1735. Burpee, R. W., and Coauthors, 1994: Real-time guidance pro- May®eld, M., L. A. Avila, and E. N. Rappaport, 1994: The Atlantic vided by NOAA's Hurricane Research Division to forecast- hurricane season of 1992. Mon. Wea. Rev., 122, 517±538. ers during Emily of 1993. Bull. Amer. Meteor. Soc., 75, Miller, B. I., 1964: A study on the ®lling of Hurricane Donna (1960) 1765±1783. over land. Mon. Wea. Rev., 92, 389±406. Charnock, H., 1955: Wind stress on a water surface. Quart. J. Roy. NHC, 1992: Hurricane Andrew Preliminary Report. 36 pp. [Avail- Meteor. Soc., 81, 639±640. able from National Hurricane Center, 11691 SW 117th St., Mi- Cline, I. M., 1926: Tropical Cyclones. The MacMillan Co., 301 pp. ami, FL 33165-2149.] Davenport, A. G., 1961: The spectrum of horizontal gustiness near OFCM, 1992: Minutes of the 46th Interdepartmental Hurricane Con- the ground in high winds. Quart. J. Roy. Meteor. Soc., 87, 192± ference. 229 pp. [Available from Of®ce of the Federal 211. Coordinator for Meteorological Services and Supporting Re- Deardorff, J. W., 1972: Parameterization of the planetary boundary search, Suite 1500, 8455 Colesville Rd., Silver Spring, MD layer for use in general circulation models. Mon. Wea. Rev., 100, 20910.] 93±106. , 1993a: Minutes of the 47th Interdepartmental Hurricane Durst, C. S., 1960: Wind speeds over short periods of time. Meteor. Conference. 406 pp. [Available from Of®ce of the Federal Mag., 89, 181±186. Coordinator for Meteorological Services and Supporting Re- Forristall, G. Z., 1988: Wind spectra and gust factors over water. search, 8455 Colesville Rd., Suite 1500, Silver Spring, MD Proc. Offshore Technology Conf., Houston, TX, OTC, 449± 460. [Available from University of Miami RSMAS Li- 20910.] brary, 4600 Rickenbacker Cswy., Miami, FL 33149-1098.] , 1993b: National hurricane operations plan. 125 pp. [Available Franklin, J. L., S. J. Lord, S. E. Feuer, and F. D. Marks Jr., 1993: from Of®ce of the Federal Coordinator for Meteorological Ser- The kinematic structure of Hurricane Gloria (1985) determined vices and Supporting Research, Suite 1500, 8455 Colesville Rd., from nested analyses of dropwindsonde and Doppler radar data. Silver Spring, MD 20910.] Mon. Wea. Rev., 121, 2433±2451. Panofsky, H. A., and J. A. Dutton, 1984: Atmospheric Turbulence. Fujita, T. T., 1963: Analytical mesometeorology: A review. Severe Wiley Interscience, 397 pp. Local Storms, Meteor. Monogr., No. 27, Amer. Meteor. Soc., Peterson, E. W., 1969: Modi®cation of mean ¯ow and turbulent 77±125. energy by a change in surface roughness under conditions Garratt, J. R., 1977: Review of drag coef®cients over oceans and of neutral stability. Quart. J. Roy. Meteor. Soc., 95, 561± continents. Mon. Wea. Rev., 105, 915±929. 575. Gill, A. E., 1968: Similarity theory and geostrophic adjustment. Powell, M. D., 1980: Evaluations of diagnostic marine boundary Quart. J. Roy. Meteor. Soc., 94, 586±588. layer models applied to hurricanes. Mon. Wea. Rev., 108, 757± Graham, H. E., and G. Hudson, 1960: Surface winds near the center 766. of hurricanes (and other cyclones). NHRP Rep. 39, 200 pp. , 1982: The transition of the Hurricane Frederic boundary layer [Available from NOAA/AOML Library, 4301 Rickenbacker wind ®eld from the open Gulf of Mexico to landfall. Mon. Wea. Cswy., Miami, FL 33149.] Rev., 110, 1912±1932. Hawkins, H. F., and D. T. Rubsam, 1968: Hurricane Hilda, 1964. II. , 1987: Changes in the low-level wind kinematic and thermo- Structure and budgets of the hurricane on October 1, 1964. Mon. dynamic structure of Hurricane Alicia (1983) at landfall. Mon. Wea. Rev., 96, 617±636. Wea. Rev., 115, 75±99. , and S. M. Imbembo, 1976: The structure of a small, intense , 1993: Wind measurement and archival under the Automated hurricaneÐInez 1966. Mon. Wea. Rev., 104, 418±442. Surface Observing System (ASOS): User concerns and oppor- Houston, S. H., and M. D. Powell, 1994: Surface winds and water tunity for improvement. Bull. Amer. Meteor. Soc., 74, 615±623. level response during Tropical Storm Marco (1990). Wea. , and P. G. Black, 1990: The relationship of hurricane recon- Forecasting, 9, 427±439. naissance ¯ight-level measurements to winds measured by Janjic, Z. I., 1994: The step-mountain eta coordinate model: NOAA's oceanic platforms. J. Wind Eng. Ind. Aerodyn., 36, Further developments of the convection, viscous sublayer, 381±392. and turbulence closure schemes. Mon. Wea. Rev., 122, 927± , and S. H. Houston, 1996: Hurricane Andrew's wind ®eld at 945. landfall in south Florida. Part II: Applications to real-time anal- Janssen, P. A. E. M., 1989: Wave-induced stress and the drag of air ysis and preliminary damage assessment. Wea. Forecasting, 11, ¯ow over sea waves. J. Phys. Oceanogr., 19, 745±754. 329±349.

/3q05 0218 Mp 327 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218 328 WEATHER AND FORECASTING VOLUME 11

, P. P. Dodge, and M. L. Black, 1991: The landfall of Hurricane van der Hoven, I., 1957: Power spectrum of horizontal wind speed Hugo in the Carolinas: Surface wind distribution. Wea. Fore- in the frequency range of 0.0007 to 900 cycles per hour. J. casting, 6, 379±399. Meteor., 14, 160±164. Sempreviva, A. M., S. E. Larsen, N. G. Mortensen, and I. Troen, Wallace, R., 1992: How fast was Andrew? Engineers, meteorologists 1990: Response of neutral boundary layers to changes of rough- disagree. Miami Herald, September 6, p. 24a. [Available from ness. Bound.-Layer Meteor., 50, 205±225. Miami Herald Publishing Co., One Herald Plaza, Miami FL Simiu, E., and R. H. Scanlan, 1986: Wind Effects on Structures. Wi- 33132-1693.] ley Interscience, 589 pp. WERC, 1992: Hurricane AndrewÐPreliminary observations of the Simpson, R. H., and H. Riehl, 1981: The Hurricane and Its Impact. WERC post disaster team. 5 pp. [Available from Wind Engi- Louisiana State University Press, 398 pp. neering Research Council, Inc., P.O. Box 10029, College Sta- Singer, I. A., C. M. Nagle, and R. M. Brown, 1961: Variation of tion, TX 77842.] wind with height during the approach and passage of Hurri- Wieringa, J., 1992: Updating the Davenport roughness classi®cation. cane Donna. NHRP Rep. 50, 408 pp. [Available from J. Wind Eng. Ind. Aerodyn., 41, 357±368. NOAA/AOML Library, 4301 Rickenbacker Cswy., Miami, Willoughby, H. E., 1988: The dynamics of the tropical cyclone core. FL 33149.] Aust. Meteor. Mag., 36, 183±191. Tuttle, J. D., and B. Gall, 1995: Radar analysis of Hurricanes , 1990: Gradient balance in tropical cyclones. J. Atmos. Sci., 47, Andrew and Hugo. Preprints, 21st Conf. on Hurricanes and 265±274. Tropical Meteorology, Miami, FL, Amer. Meteor. Soc., 608± Wood, D. H., 1982: Internal boundary layer growth following a step 610. change in surface roughness. Bound.-Layer Meteor., 22, 241±244.

/3q05 0218 Mp 328 Wednesday Aug 07 05:51 PM AMS: Forecasting (September 96) 0218